app.py

import gradio as gr
import tensorflow as tf
import numpy as np
import cv2
import os
import requests
from tensorflow.keras.models import load_model
from PIL import Image, ImageDraw

# DEBUG: Print start message
print("Starting Seamount Detection App...")

# Define model path and URL for dynamic downloading
MODEL_PATH = "objdet_1_2.h5"
MODEL_URL = "https://huggingface.co/YOUR_USERNAME/objdet_1_2/resolve/main/objdet_1_2.h5"

# Download model if not present
if not os.path.exists(MODEL_PATH):
    print("Model not found. Downloading from:", MODEL_URL)
    response = requests.get(MODEL_URL)
    with open(MODEL_PATH, 'wb') as f:
        f.write(response.content)
    print("Model downloaded successfully.")
else:
    print("Model found locally.")

# Load the model
print("Loading model...")
model = load_model(MODEL_PATH, custom_objects={'huber_loss': tf.keras.losses.Huber()})
print("Model loaded.")

# Constants
IMAGE_SIZE = (256, 256)

# Preprocessing function
def preprocess_image(image):
    print("Preprocessing image...")
    image = image.convert("RGB")
    image = np.array(image) / 255.0
    image = cv2.resize(image, IMAGE_SIZE)
    return np.expand_dims(image, axis=0)

# Prediction function
def predict(image):
    print("Running prediction...")
    processed_image = preprocess_image(image)
    pred_class, pred_bbox = model.predict(processed_image)
    confidence = float(pred_class[0][0])
    classification = "Seamount" if confidence > 0.5 else "No Seamount"
    width, height = image.size
    bbox = pred_bbox[0]
    x_min = int(bbox[0] * width)
    y_min = int(bbox[1] * height)
    x_max = int(bbox[2] * width)
    y_max = int(bbox[3] * height)
    print(f"Prediction: {classification} with confidence {confidence:.2f}")
    return (
        classification, 
        confidence, 
        (x_min, y_min, x_max, y_max), 
        draw_bounding_box(image, (x_min, y_min, x_max, y_max), classification)
    )

# Function to draw bounding box on the image
def draw_bounding_box(image, bbox, label):
    print("Drawing bounding box...")
    x_min, y_min, x_max, y_max = bbox
    image = image.convert("RGB")
    draw = ImageDraw.Draw(image)
    draw.rectangle([x_min, y_min, x_max, y_max], outline="red", width=3)
    text_position = (x_min, y_min - 10)
    draw.text(text_position, label, fill="red")
    return image

# List of example image file paths (ensure these images are uploaded to your Space)
example_image_paths = [
    "example1.png",
    "example2.png",
    "example3.png",
    "example4.png"
]

# Build the Gradio interface using Blocks
with gr.Blocks() as demo:
    gr.Markdown("# Seamount Detection")
    gr.Markdown("**Instructions:** Either upload your own image or select one of the example images below. Clicking an example will populate the input field.")

    with gr.Row():
        # Single image input component
        uploaded_image = gr.Image(type="pil", label="Upload an Image")
    
    # Examples component displays the images as clickable thumbnails.
    gr.Examples(
        examples=[[path] for path in example_image_paths],
        inputs=uploaded_image,
        label="Or select an example image"
    )
    
    submit_btn = gr.Button("Predict")
    
    with gr.Row():
        output_class = gr.Text(label="Classification")
        output_confidence = gr.Text(label="Confidence Score")
        output_bbox = gr.Text(label="Bounding Box (x_min, y_min, x_max, y_max)")
    
    output_image = gr.Image(label="Image with Bounding Box")
    
    submit_btn.click(
        predict,
        inputs=uploaded_image,
        outputs=[output_class, output_confidence, output_bbox, output_image]
    )

# DEBUG: Launch the app
print("Launching app...")
if __name__ == "__main__":
    demo.launch()